The rapid advancement and adoption of artificial intelligence in recent years can be largely attributed to the open sharing of knowledge and transparent collaboration within a growing community. By fostering an environment that encourages debate, criticism and ideas, professionals can build upon the work of others and drive innovation forward. Instead of reinventing the wheel each time, new solutions benefit from the foundation laid by previous efforts.

Researchers can adopt different degrees of open knowledge sharing when publishing their work. These approaches can range from simply sharing a paper with a description of the experiments and results obtained, to releasing the source code behind the project, opening the data sets to the public, or even publishing the model definition and internal weights for others to use. The level of openness chosen can significantly impact the level of collaboration and innovation within the community.

As a recent example, in the field of text-to-image generation, the OpenAI team published a state-of-the-art paper about DALL-E 2 and made it available for commercial paid use. On the other hand, StabilityAI released Stable Diffusion model under an open license. Anyone could access and use the code and data for Stable Diffusion, even retraining the model for their own purposes. The fully open approach of StabilityAI has led to its speedy adoption and the development of numerous applications. Although this example might be an extreme case, it demonstrates the impact of shared knowledge combined with a radical openness to drive innovation.

The air traffic management (ATM) industry has shown a different trend than the AI community. The sector is composed mainly of large players who often struggle to innovate, likely because they have less incentive to do so. On the other hand, small companies and research institutes eager to find disruptive solutions often face high barriers to accessing data and resources. This scenario, along with the safety-critical nature of the ATM industry, make the adoption of AI solutions slow: the ATM also has a heavily regulated landscape, significantly smaller size of the sector compared to other industries, and significant uncertainty about the roadmap and certification of ML and AI systems. We can do better, don’t you think? A more open and transparent community can only accelerate the shift towards greater automation and more efficient and safer air traffic management that provides a scalable response to the growth in demand for navigation services.

Could public institutions facilitate access to data in a safe, more agile way? With fewer bureaucratic barriers for companies to develop innovative solutions? Should we be satisfied with how the results of projects financed by European funds are presented in a series of reports, published articles or posters? Or can we go further and make the source code, data or models available to the community? In short, can we increase the transparency and accessibility of data and knowledge in the industry while respecting the protection of confidential and proprietary information?

There are reasons for optimism. The open community is growing. There are more and more public and private initiatives with a more open philosophy, as well as individual projects and initiatives that are born with the purpose of sharing knowledge. The following is a non-exhaustive list of examples:

• OpenSky Network. A non-profit association based in Switzerland. It aims to improve air space usage’s security, reliability and efficiency by providing public access to real-world air traffic control data.
• Traffic. A toolbox for processing and analysing air traffic data, being developed mainly by Xavier Olive. The library also offers facilities to parse or access traffic data from open sources of ADS-B traffic like the OpenSky Network or Eurocontrol DDR files, easily extendable to other data sources.
• BlueSky. An open-source air traffic simulator developed by researchers from TU-Delft. The purpose of BlueSky is to offer a freely accessible tool for anyone interested in visualising, analysing, or simulating air traffic without any restrictions, licenses, or limitations.
• PyModeS. PyModeS is a Python library designed to decode Mode-S (including ADS-B) messages created by Junzi Sun, a researcher from TU-Delft.
• TrajAirNet. A project from Carnegie Mellon University’s Air lab. They release a dataset from recorded trajectories of aircraft operating around a standard non-towered airport while also providing the weather conditions during these operations and the source code for a socially aware trajectory prediction model.

To sum up, embracing a more open and transparent approach to collaboration will at least speed up innovation. As a result, we can build upon the successes of brilliant researchers and practitioners and ultimately create a brighter future for AI in the ATM industry.